I am training a semantic segmentation network (I'm trying SegNet) to classify skin in an image. This is the only class that will need to be labeled and I labeled all my training and testing data using the Image Labeling App in Matlab, so if you load an image you will see the background as zeros and skin as ones.
However, I didn't specify a 'background' class as that seemed unnecessary, I just didn't label it. So when I try to build a network with one class (calling lgraph = segnetLayers(imageSize,numClasses,model) ) using one as the number of classes it says the number of classes needs to be greater then one. Is this just an attribute of these types of networks that you need to have two or more classes for the network to classify? If I set it to two it obviously fails during training but I'm also not sure if labeling everything else as 'background' will yield positive results (I'm thinking it may just classify everything as background). Thoughts?